Racing Vehicle Control Using Insect Intelligence

1
Racing Vehicle Control Using Insect Intelligence

• Ivar Mossin Ralph Meijer

2
Insect Intelligence

• What is it?
• Artificial Life Methods
• Hierarchical Decision-making
• Simple toolset of sensors, functions and
  activators

3
Insect Intelligence

• Seldom used to directly control game entity
• Restricted minimal representation
• Physical model what the agent thinks
• A combination of three things
• Exteroception
• Proprioception
• Body sense

4
Insect Intelligence

• Implementation
• On each iteration
• Updates all agents with sensor inputs
• Each agent express its behavior
• Behavior of the agent is read from output
• Game updates physical agent-state

5
Insect Intelligence

• Traditional closed-loop feedback system
• Measurement compared with setpoint
• Decides action to compensate for error
• Insect Intelligence approach
• Sensor-inputs compared with agent wants
• Decides action to reach goal

6
Insect Intelligence

• What is it good for?
• Many simple tasks instead of one complex
• Easier to understand, design and implement
• Performs complex tasks when put together
• Self-calibrating
• Differentiable / nondifferentiable operations

7
Insect Intelligence

• Integrating with a Game Engine
• Does not need direct access to the games physics
  engine
• Can simulate user-input

8
An Insect Intelligent Inspired Application

• The Racing game Downforce

9
The Fundamental Data Abstraction

• Track Space
• Transforms the position of the car
• Removal of corners
• Representing the track as a long strip
• Width 1.0
• (Infinite) Length
• fWidthPos
• fLengthPos

10
Implementation Track Space Transform

• Array of Track lines
• Left Edges, Right Edges, and Race lines
• fLengthPos value
• The whole number component
• The fractional component

11
Implementation Track Space Transform
12
Downforce AI system overview

• Layered approach derived from Brooks
  Subsumption Architecture
• Downforce AI is primarily reactive
• Hierarchically organized layers
• Responsible for specific tasks

13
Downforce AI system overview
14
Downforce AI system overview

• Behaviours
• Layers are composed into sub-tasks or behaviours
• Interaction between layers
• Override the output
• Affect the internal state of another layer
• Change another layers input

15
Path Following Layer

• Responsible for following a path at maximized
  speed
• Steering control behaviour
• Speed regulation behaviour

16
Tactical Racing Layer

• Responsible for making tactical decisions
• Choose most appropriate road position
• Generating a new racing line
• Collision Avoidance speed control

17
Fine grain avoidance layer

• Responsible to deal with potential collision
  situations
• Race priority behaviour
• Collision avoidance behaviour

18
Driving Assist Layer

• Responsible for control
• Maximum traction under acceleration
• Preventing the brakes from locking

19
Conclusions of the AI application Downforce

• Its behavioural functionality is modular,
  extensible, and scalable
• Complexity is reduced through decentralization

20
References

• Biasillo, Gari, Racing AI Logic. AI Game
  Programming Wisdom, Electronic Arts Canada, 2002.
• Darby, Alex, Racing Vehicle Control using Insect
  Intelligence. AI Game Programming Wisdom 2,
  FreeStyleGames Ltd., 2003.
• Porcino, Nick, Insect AI 2 Implementation
  Strategies. AI Game Programming Wisdom 3,
  LucasArts, a Lucasfilm Company, 2006

21

• Questions